Upload
others
View
13
Download
0
Embed Size (px)
Citation preview
Dynamic Power-Aware Shared Path Protection Algorithms
in WDM Mesh Networks
Rongxi He1, 2 and Bin Lin1
1. College of Information Science and Technology, Dalian Maritime University, Dalian 116026, China
2. Key Lab of Optical Fiber Sensing & Communications (UESTC), Ministry of Education, Chengdu 611731, China
Email: [email protected]; [email protected]
Abstract—In this paper, we investigate the dynamic
establishment of dependable connections in wavelength
division multiplexing (WDM) mesh networks with the
objectives of power saving and resource efficiency.
According to the resource usage, i.e., assigned for primary
paths, backup paths or free, and the state of network
components (nodes and fiber links), i.e., active, sleep or off,
we propose a dynamic power-aware shared path protection
(DPA-SPP) algorithm. In order to reduce power
consumption, to improve sharing of spare capacity, and to
reduce blocking probability for connection requests, DPA-
SPP encourages to establish a primary path for each
connection request on active resources, to pack backup
paths on sleep resources, and to leave more idle resources
for future connection requests as far as possible while with a
consideration to prevent a link with a very few number of
free wavelengths to become a bottleneck link, which is
beneficial to further improve the successful probability of
connection establishments. Based on dynamic traffic with
different load, the performance of DPA-SPP has been
investigated via extensive simulations. The results show that
DPA-SPP can efficiently improve the spare resource sharing
and reduce the blocking probability while still achieving a
considerable energy saving.
Index Terms—Green networks; energy saving; shared path
protection; spare capacity; dynamic
I. INTRODUCTION
Due to the rapid growth of broadband users and
emerging bandwidth-intensive applications, such as video
on demand (VoD), high-definition television (HDTV),
interactive gaming, etc., the traffic supported by the
Internet has scaled up significantly in the last decade [1]-
[3]. In optical networks employing wavelength-division-
multiplexing (WDM) technology, hundreds of
independent wavelength channels can be multiplexed
along a single fiber, each with a transmission rate
exceeding 10Gb/s or even 40 Gb/s [4]. With equivalent to
Tb/s effective bandwidth, WDM networks have been
playing a key role in the backbone networks and Internet.
However, with the increase of transmission rate and
capacity, more and more equipment and components are
required to deploy globally, and the energy consumption
is also grown at a high rate. If the appetite for energy
continues to scale up rapidly, energy shortage will
Manuscript received July 1, 2012; revised August 1, 2012; accepted
September 1, 2012.
hamper the expansion of future Internet [3]-[7]. Therefore,
it is imperative to develop energy-efficient backbone
networks.
Recently, many research efforts have been focused on
designing energy-efficient (or power-efficient) WDM
optical networks as environmental-friendly solutions for
backbone networks. The work in [1] provided a
comprehensive survey of energy-conservation protocols
and energy-efficient architectures over different domains
of telecom networks. A power consumption model for
transparent circuit-switched WDM networks has been
provided in [2]. Based on the model, an ILP formulation
with the objective to minimize the power consumption
has been given and a simple heuristic algorithm has also
been proposed to solve it. In the pioneering work of [5],
the authors introduced the power-aware routing and
wavelength assignment (PA-RWA) problem and gave an
ILP-based formulation with the goal to minimize the
power consumption. A heuristic algorithm has also been
presented to solve it under a static traffic demand.
Concentrating on minimizing the energy consumption for
an IP over WDM network, an MILP optimization model
and efficient heuristics with the lightpath bypass strategy
have been developed in [6]. In [7], a network-based
power-consumption model, taking into account energy
consumption in switching and transmission equipment,
has been proposed for optical IP networks to estimate the
energy consumption of the Internet.
However, the aforementioned work mainly focused on
the static design problem for energy-efficient WDM
networks. A dynamic PA-RWA algorithm has been
proposed in [8] to improve the energy efficiency of
WDM networks by packing traffic on as few links as
possible to minimize the number of optical amplifiers
being active in the networks. As an extension of [8], an
ILP formulation for the static lightpath establishment and
a few heuristics for the dynamic lightpath establishment
with the consideration of power consumption in multi-
fiber WDM networks have been proposed in [9]. The
author in [10] provided an energy-efficient dynamic
connection provisioning scheme for WDM networks,
based on an intelligent load control mechanism and an
auxiliary model. In [11] and [12], the authors investigated
the adverse effect on network blocking probability
performance of PA-RWA solution, and proposed a
weighted approach to make a trade-off between power
saving and blocking probability. The authors in [3]
Journal of Communications, Vol. 8, No.1, January 2013
55©2013 Engineering and Technology Publishing doi: 10.12720/jcm.8.1.55-65
investigated green provisioning strategies for traffic-
grooming WDM networks from the component layer to
network layer, and developed a power-aware scheme to
minimize the total operational power.
A single fiber failure can cause the failures of all the
lightpaths passing through the fiber, which leads to the
loss of huge amount of data and revenue. Therefore,
effective survivability mechanisms are stringent
requested to minimize the data loss in WDM networks
[13], [14]. There are two ways against single fiber failure
in optical layer, i.e., dynamic restoration and preplanned
protection [13], [14]. Protection, reserving spare capacity
during lightpath setup, is essential for recovering from
failures in a short time period. It has two main forms:
shared protection (SP) and dedicated protection (DP).
Conventionally, SP can broadly be classified into shared
link-protection (SLP) and shared path-protection (SPP)
[13, 14]. For each connection request, SPP first sets up a
primary path, and then reserves a fiber-disjoint backup
path and wavelengths for it. The reserved resources on
one backup path can be shared among other backup paths
if their corresponding primary paths are fiber-disjoint.
Conventional SPP only focuses on improving the spare
capacity sharing without considering the energy saving.
In order to efficiently utilize the network resources and to
improve the energy efficiency, it is better to consider the
two problems simultaneously.
In [15], the authors investigated the energy-efficient
network planning problem for resilient WDM networks
that exploits the sleep mode of devices with dedicated
path protection, and formalized it as an ILP formulation.
Under the assumption that all lightpath requests are
known in advance, an ILP-based formulation and a
heuristic algorithm for the energy-efficient shared backup
protection in WDM networks have been developed in [16]
and [17], respectively. None of the above work has
involved in the energy-aware survivable connection
provisioning problem in a dynamic scenario. The authors
of [4] proposed and evaluated different energy-efficient
algorithms to dynamically provision 1:1 dedicated path
protected connection for WDM networks with a sleep
mode option. In [18], the authors proposed a sleep mode
based power-aware SPP algorithm to achieve the power
efficiency of WDM networks. However, in order to
increase the load carried in active components and to turn
off (or switch into sleep) as many network components as
possible, the aforementioned power-aware strategies,
with an overemphasis on power saving, tend to pack
primary paths and backup paths in active links and sleep
links respectively without considering resource efficiency.
Potentially, they may choose a long path with more hops
and lead to a high possibility of overutilization for some
links in the network, particularly with an increase of
network load. More hops mean more resources being
occupied and more links being overutilization mean more
possibility to lead to a temporary division in the network,
which results in a less chance to successfully establish a
dependable path for future connection requests. On the
other side, the conventional SPP, i.e., power-unaware
shared path protection (PU-SPP), with the objective to
improve the utilization of spare resources, makes a great
effort to reduce the resources occupied by backup paths
while with an overlook to avoid more sleep components
to be activated for energy saving. In order to reduce
energy consumption, it is much better for SPP to jointly
consider power saving and improvement of network
performance in term of traditional metrics such as
blocking probability and resource utilization. Moreover,
most of the previous power-aware strategies only
consider the energy consumption of fiber links but
omitting to reduce the power consumption of nodes.
Accordingly, a more general case for reducing power
consumption should pick a route with less number of
links and nodes in sleep or off state for a primary path
and a route with less number of links and nodes in active
state for a backup path.
The main objective of this paper is to investigate
energy efficient shared path protection under a dynamic
scenario, mainly focusing on the reduction for power
consumption of nodes and links, the improvement for
sharing of spare capacity and the prevention of link
overutilization to reduce blocking probability of
connection requests. We first introduce bottleneck link to
indicate a link whether or not being overutilization, and
then propose a dynamic power-aware shared path
protection (DPA-SPP) algorithm for WDM mesh
networks to jointly consider power saving and resource
utilization. Our work differs from previous work [17], [18]
in that we not only concentrate on power-saving for SPP
with a joint consideration for the power consumptions of
both nodes and links, but also endeavor to achieve a
global efficiency in resource allocation by improving the
share ratio of spare capacity and to prevent links being
overutilization. The latter is favorable to further reduce
blocking probability.
The rest of this paper is organized as follows. Section
II presents the network model and describes the intuition
of DPA-SPP by comparing its operation with two other
shared path protection algorithms with different
objectives. Section III describes the proposed algorithm.
Simulation results are presented in Section IV.
Conclusions follow in Section V.
II. SYSTEM MODEL AND PROBLEM ANALYSIS
A. Network Model
Define a network topology G(N, L, W) for a given
WDM mesh network, where N is the set of nodes, L is the
set of bidirectional links, and W is the set of available
wavelengths per fiber. |N|, |L| and |W| denote the number
of nodes, links and wavelengths, respectively. Each node
contains an electronic control system (ECS), a 3D Micro
Electromechanical System (MEMS) based optical
switching matrix, enough tunable transceivers, and a pair
of passive optical MUX/DEMUX for each line interface
[8],[9],[18].
Journal of Communications, Vol. 8, No.1, January 2013
56
1
2
5
7
4
3
6
8
primary path backup path
p1
p2p3
b2
Figure 1. Comparison of different shared path protection algorithms for 6 connection requests r1(1, 4), r2(1, 6),
r3(1, 7), r4(6, 8), r5(7, 8) and r6(2, 5): (a) a power-unaware SPP (PU-SPP) mainly focusing on spare resource
sharing; (b) a conventional power-aware SPP (PA-SPP) mainly focusing on energy saving; (c) a power-aware
SPP with a joint consideration of energy efficiency and network performance (DPA-SPP).
(a) (b)
b1b6
1
2
5
7
4
3
6
8
p1
p6
b2
b1
p2
1
2
5
7
4
3
6
8
p1
p2
p6
b2
(c)
b1b6
p6
b3
p3
b3 p3
b3p4b4
p4b4
p4b4
assign the same wavelength
p5b5 p5
b5 p5
b5
All network components (nodes or fiber links) have
three different modes, i.e., active, sleep and off [4]. Any
component included in a primary path is in active mode.
An active component has a full function with a certain
amount of power consumption. Any component only
reserved for backup paths (i.e., not used by any primary
path) is in sleep mode, which consumes a negligible
amount of power and can be promptly activated if needed.
In off mode, a component is not in an operation state (i.e.,
it is idle and its resources are not used.) and consumes no
power [4].
Without loss of generality, we assume that all nodes
have enough interfaces to process all the traffic that can
potentially flow through it and have full wavelength
conversion capabilities. Connection requests arrive at the
network dynamically with no knowledge of future
arrivals, and there is only one connection request with a
bandwidth requirement of one wavelength unit arriving at
the network at a time, defined by r(s, d), where s, dN
denote the source node and the destination node of the
connection request.
B. Problem Statement
To enhance power saving, the aforementioned power-
aware SPP strategies (PA-SPP) deliberately pack primary
paths and backup paths in different fiber links as far as
possible regardless of resource utilization and link
overutilization. On the other side, the conventional
power-unaware SPP strategies (PU-SPP) only emphasize
the improvement of the sharing of spare resources with a
disregard for energy saving. Correspondingly, PA-SPP
has higher resources occupation and higher possibility for
link overutilization, which results in higher blocking
probability and lower resource utilization, and PU-SPP
has higher energy consumption.
Fig. 1 shows an example to compare PA-SPP and PU-
SPP with a power-aware shared path protection (DPA-
SPP) considering the trade-off between power
consumption and resource usage, where all the links
represent bidirectional fibers and have the same physical
length and each fiber have two wavelengths. Let r1(1, 4),
r2(1, 6), r3(1, 7), r4(6, 8), r5(7, 8) and r6(2, 5) be the first
six connection requests for the specified source and
destination node pairs. For the first request r1, due to all
the resources being free, both PU-SPP (Fig. 1 (a)) and PA-
SPP (Fig. 1 (b)) choose route p1(1-2-3-4) as the primary
path and a link-disjoint route b1(1-5-6-4) as the backup
path. The connection occupies one wavelength in each
link along p1 for traffic transmission and reserves one
wavelength in each link of b1 as spare resource to
maintain service continuity against a single link failure.
When r2 arrives, PU-SPP, without differentiating the
links with resources that are free (in off mode), occupied
by primary paths or reserved for backup paths, mainly
focuses on improving the resource utilization, and picks
the shortest route p2(1-5-6) as the primary path and a
link-disjoint path b2(1-7-8-6) as the backup path. It uses
one wavelength in each fiber along p2 and reserves one
wavelength in each fiber along b2. On the other hand, the
PA-SPP with the objective to minimize the power
consumption by turning off idle components or by
switching components only reserved for backup paths
into sleep mode, deliberately chooses the link that has
already been included in primary paths for previously
established connections as the primary path, and packs
backup path on network components included in backup
paths for previous connections. As a result, it picks route
p2(1-2-3-6) as the primary path and route b2(1-5-6) as the
backup path, and occupies one wavelength in each fiber
along p2 and reserves a wavelength in each fiber along b2.
When r3 arrives, there is no difference between PU-SPP
and PA-SPP. Both of them pick the shortest route p3(1-7)
as the primary path and a link-disjoint path b3(1-5-7) as
the backup path. They use one wavelength in each fiber
along p3 and share the wavelength in the fiber along 1-5
previously reserved for r1 due to no common link
between p1 and p3 corresponding to requests r1 and r3 and
reserve a wavelength in the link along 5-7. When r4
arrives, there is also no difference between PU-SPP and
PA-SPP. Both of them pick the shortest route p4(6-8) as
the primary path and a link-disjoint path b4(6-4-8) as the
backup path. They use one wavelength in each fiber
along p4 and share the wavelength in the fiber along 6-4
previously reserved for r1 and reserve a wavelength in
each link along 4-8. For r5, PU-SPP and PA-SPP pick
route p5(7-8) as the primary path and a link-disjoint path
Journal of Communications, Vol. 8, No.1, January 2013
57
b5(7-5-6-8) as the backup path and use one wavelength in
each fiber link along p5. PU-SPP shares the wavelength in
the fiber along 7-5, 5-6 and 6-8 previously reserved for r3,
r1 and r2, respectively. PA-SPP shares the wavelength in
the fiber along 7-5 and 5-6 previously reserved for r3 and
r2, and reserves a new wavelength in the fiber along 6-8.
Since PA-SPP only considers energy saving by packing
more primary paths on active components and more
backup paths on components in sleep mode without
considering the sharing of spare resource, it may increase
the unbalance of resource usage and lead to more links
without free resource for future requests. Therefore, it is
more possible to increase the blocking probability for
future connections. For example, when a new request r6(2,
5) arrives in Fig. 1, PU-SPP can choose p6(2-5) as the
primary path and b6(2-1-5) as the backup path. It uses one
wavelength in the fiber along p6 and reserves one
wavelength in the fiber along 2-1 and shares one
wavelength in the fiber along 1-5 previously reserved for
r1. However, PA-SPP can choose p6(2-5) as the primary
path, but cannot find a link-disjoint backup path with
available spare capacity, since it strives hard to pack
primary paths and backup paths on active and sleep
components respectively without considering the
improvement of resource utilization and the prevention of
link overutilization.
In order to accept more connection requests and to
improve the resource utilization, it is better for a dynamic
SPP to jointly consider power-saving and spare capacity
sharing and to keep links from overutilization as much as
possible. The main objective of this paper is to propose
such a dynamic power-aware SPP (DPA-SPP). Fig. 1 (c)
gives an example for the operation of DPA-SPP for the
above six connection requests. For the first request r1,
being the same as PU-SPP and PA-SPP, DPA-SPP
chooses route p1(1-2-3-4) as the primary path and b1(1-5-
6-4) as backup path, and occupies one wavelength in each
fiber along p1 for traffic transmission and reserves one
wavelength in each link of b1 as the spare resource. For r2,
with a great effort to pack primary path and backup path
on different links and to avoid using up the wavelengths
in links (1, 2) and (2, 3) or (1, 5) and (5, 6), DPA-SPP
picks route p2(1-7-8) as the primary path instead of (1-2-3)
being picked by PA-SPP and (1-5-6) being picked by PU-
SPP. It assigns one wavelength in each link along p2.
Correspondingly, b2(1-5-6) is chosen as backup path and
it shares the wavelength in links (1, 5) and (5, 6) reserved
for b1, due to no common link involved in p1 and p2. For
requests r3 and r4, DPA-SPP picks the shortest route p3(1-
7) and p4(6-8) as the primary path and a link-disjoint path
b3(1-5-7) and b4(6-4-8) as the backup path for r3 and r4,
respectively. It uses one wavelength in each fiber along p3
and p4 and reserves one wavelength in the fiber along b3
and 4-8 and shares the wavelength in the fiber along 6-4
previously reserved for r1. For request r5, it picks route
p5(7-8) as the primary path and a link-disjoint path b5(7-
5-6-4-8) as the backup path. It uses one wavelength in
each fiber along p5 and reserves a new wavelength in the
fiber along 5-6 and shares the wavelength in the fiber
along 7-5, 6-4 and 4-8 previously reserved for r3, r1 and
r4, respectively. When r6 arrives, it picks p6(2-5) as the
primary path and a link-disjoint route b6(2-3-6-5) as the
backup path. It uses one wavelength in the fiber along p5
and reserves one wavelength in each fiber along (2-3-6)
and shares the wavelength in the fiber (6-5) previously
reserved for r1.
Table I presents the link usage of PU-SPP, PA-SPP
and DPA-SPP shown in Fig. 1. Considering the
negligible power consumption of components in sleep or
off (idle) mode compared with active mode, it is clearly
shown from Table I that an power-aware approach, i.e.,
PU-SPP or DPA-SPP, has a better energy saving. Table I
shows that the number of active links is 7 for PA-SPP and
DPA-SPP, and the number of links in sleep or off mode is
6. On the other hand, PU-SPP has 9 links in active mode
and only 4 links in sleep or off mode. Obviously, with the
objective to minimize the number of active components,
power-aware strategies can achieve more energy saving.
Although only 5 connections can be successfully
established by PA-SPP for the 6 arriving requests in Fig.
1 (b), there is 5 links without available wavelength for
future requests, i.e., (1, 2), (2, 3), (1, 5), (5, 6) and (6, 8).
On the other side, DPA-SPP not only considers power
saving but also aims at the improvement of resource
sharing and the prevention of link overutilization. It also
has 5 links without available wavelengths even if it
successfully accepts all the 6 connection requests.
Potentially, compared with PA-SPP, DPA-SPP has a
higher chance to accept more future connection requests,
which leads to a smaller blocking probability and a higher
resource utilization.
III. DESCRIPTION OF THE PROPOSED ALGORITHM
The critical issue addressed in this paper is how to
select a primary path and a link-disjoint backup path with
the objectives to improve spare capacity sharing and to
reduce energy consumption for a dynamic connection
request. As mentioned above, PA-SPP tends to select the
path with more active components as primary path and
the path with more sleep components as backup path,
although those paths are not the shortest path while using
other metrics (such as hop number and resource
utilization, etc.). It may increase the average path length
and occupy more resources in the network. To deal with
this problem, a new dynamic power-aware SPP (called
DPA-SPP) will be proposed in this paper with a
TABLE I. LINK USAGE FOR PU-SPP, PA-SPP AND DPA-SPP
Algorithm
Successfully
Accepted
requests
Number of links
Active Sleep Off
PU-SPP 6 9 3 1
PA-SPP 5 7 5 1
DPA-
SPP 6 7 6 0
Journal of Communications, Vol. 8, No.1, January 2013
58
comprehensive consideration of power saving and
improvement of resource utilization and network
performance.
Upon the arrival of connection request r(s, d), DPA-
SPP first computes a primary path and assigns one
wavelength in each fiber along the path. And then, it
picks a fiber-disjoint path as the backup path. Since spare
resource can be shared among different backup paths
whose corresponding primary paths are fiber-disjoint, it is
unnecessary to reserve one wavelength in all backup
paths [14]. The amount of the reserved wavelength is
determined by the link-usage information that is available
to the routing algorithm. This incremental information is
feasible to obtain from traffic engineering extensions to
routing protocols [19]. Before describing the proposed
scheme, following notations are introduced.
(i, j): a fiber link interconnecting node i and node j in
the topology G, which represents two unidirectional links
between the two nodes.
cij: the cost of (i, j), which is determined by the
physical topology and the current state of the network.
P: the set of links passed by any primary path.
B: the set of links involved in any backup path.
aij: the total number of wavelengths used by primary
paths in link (i, j).
rij: the total number of wavelengths reserved for
backup paths in link (i, j).
fij: the total number of residual wavelength in link (i, j).
Obviously, fij=|W|−aij−rij.
xij: a binary variable that is equal to 1 if ijf W ,
and equal to 0 otherwise, where α is a weighting factor
considering the prevention of link overutilization, and
0<α<0.5. If xij=1, (i, j) is called a bottleneck link, which
means that (i, j) has a very few free wavelengths for
future connection requests and is overutilization. For a
bottleneck link, DPA-SPP encourages to pick another
link instead of it to create the connection for the arriving
request, in order to avoid using up its resource and
potentially leading to a division of the network. However,
in a conventional power-aware SPP (PA-SPP), no
attention is paid to prevent link overutilization. It only
stresses on packing primary path onto active components
and putting more components into sleep or off mode.
zij: a binary variable that is equal to 0 if (i, j)P, and
equal to 0 otherwise.
ei: a binary variable that is equal to 0 if node i is used
by any primary path (node i is in active mode), and equal
to 1 otherwise, i.e., node i is in off or sleep mode.
dij: the physical distance between node i and node j in
km, where i, jN.
d0: the length of a single mode fiber span, which is
fixed to 80km [4].
PE: the power consumed by an ECS;
PM: the power consumed by an optical wavelength
converter and 3D MEMs-based switching matrix per
wavelength in a node.
PT: the power consumed by a transceiver.
PILA: the power consumed by an optical in-line
amplifier (ILA).
PPRE: the power consumed by a pre-amplifier.
PPOST: the power consumed by a post-amplifier.
Pij: the power consumed by amplifiers on link (i, j)L,
which can be modeled as follows.
0
ijij ILA PRE POST
dP P P P
d
(1)
DPA-SPP first computes the K shortest paths as the
candidates for the primary path of the current connection
request. Because a link without available wavelength
cannot be used by the primary path, it should be
temporarily removed from G, i.e., modifying the cost of
edges as infinity, before computing the primary path.
Moreover, in order to reduce the blocking probability of
connection requests, to prevent link overutilization and to
enhance energy saving as much as possible, DPA-SPP
dynamically adjusts the costs of the edges according to
the current state of the network. Before computing the
primary path, DPA-SPP adjusts the cost of each link in G
according to following equations.
, 0
,
ij
ij
ij
if fc
u otherwise
(2)
( ) 2 / 2,
2 ( / ),
( , ) ( , )
( ) 2 ( / ) ,
( , ) ( , )
2 ( / ) / 4,
( , ) ( , )
i j E M ij ij
M ij max ij
i j E M ij ij max ijij
M ij max ij
e e P P P Q if f W
P x P f
if i j P and i j B
e e P P P x P f Qu
if i j B and i j P
P x P f Q
if i j P and i j B
(3)
where Pmax is a constant factor with a big value to
prevent link overutilization by encouraging to choose the
links with more free wavelengths instead of bottleneck
links. Pmax is defined as the power consumed by
amplifiers on the link with the longest distance, which
can be written as follows.
ijLji
max PmaxP
),(
(4)
In (3), Q is a constant with a very big value as a
penalty to discourage picking a specific link, which is
defined as follows.
maxPLQ (5)
In (3), Q, Q/2, and Q/4 are used to control the
preference to select different links belong to different sets.
According to Eq. (3), DPA-SPP first prefers the link only
involved in primary paths for previous connections to
establish a primary path. The cost for a link solely used
by primary paths is set to the additional energy
consumption for the link being used by current
connection plus a factor for load balance. And then it
encourages choosing the link used by primary paths and
backup paths of previous connections, with a bigger cost
than the previous case. Furthermore, in order to maximize
the number of links that can be turned off, a big cost is
Journal of Communications, Vol. 8, No.1, January 2013
59
assigned to a link being unused, that is, a constant Q/2
plus the power consumption of the link being activated
from off state, which is much bigger than the cost of the
link under above two cases. In order to maximize the
number of links that can be put in sleep mode, it is
discouraged to provisioning primary paths with resources
already reserved mainly for protection purposes.
Therefore, a very high cost is assigned to the links used
only by backup paths, which is bigger than those of the
above three cases.
After modifying the link cost, DPA-SPP computes K
shortest paths (if there are no K distinct path, then it
should find all possible results) between source node s
and destination node d using Yen’s K-shortest path
algorithm presented in [20]. All found paths make up of a
set, denoted by SK. In fact, the paths in SK are the first K
most energy-efficient paths of all potential paths between
the source and destination node with a consideration for
reduction of bottleneck links. If SK=NULL, which means
that no primary path is available for the connection
request, the request is blocked immediately. Otherwise, a
link-disjoint backup path will be computed for each path
in the set SK one by one, also with the consideration of
energy saving and improvement of resource usage. Since
spare capacity can be shared by connections being not
simultaneously failure to improve the resource utilization,
before computing a backup path for the chosen primary
path Pl from SK, modify the cost of link in G according to
following equations.
, ( , )
,
l
ij
ij ij
if i j Pc
m u otherwise
(6)
where mij and uij are dynamic cost weights for (i, j)
considering spare capacity sharing and energy saving,
respectively, which are defined as follows.
, 1
min{1, 1 }, 1
ij ij
ij ij ij ij ij ijij
if q r
q r if r q r fm
otherwise
+ ,
(7)
/ 4, if
( / ) , if ( , ) and ( , )
( / ), if ( , ) and ( , )
( / ) / 2, if ( , ) and ( , )
ij
ij max ij
ij
ij max ij
ij max ij
Q f W
x P f Q i j P i j Bu
x P f i j B i j P
x P f Q i j P i j B
(8)
where ξ is a small number being very close to 0. qij is the
maximum number of wavelength needed on link (i, j) if
any of the links along Pl fails, which can be computed
based on the network state before the new connection is
routed [19]. Equation (7) encourages picking the link
with the least amount of additional bandwidth to set up
the backup path. Equation (8) makes a deliberate attempt
to pack as many backup paths as possible on sleep
components and to discourage the use of links involved in
previous primary paths for backup paths. A very small
cost is set to the link only involved in previous backup
paths, and a highest cost is set to the link only included in
primary paths. Accordingly, (6) always select the links
with the least amount of additional bandwidth to set up
backup paths while with a consideration of energy saving.
It is favorable to improve the bandwidth utilization and to
reduce the energy consumption.
Once modifying the costs of all links, a shortest path
algorithm (e.g., Dijkstra’s algorithm [14]) is used to
compute a distinct route with minimum weight for Pl,
denoted by Pc. If it fails to find a path, DPA-SPP picks up
next path from SK and repeats above procedure.
Otherwise, put Pl and Pc into a set X, and pick up next
path from SK and repeat above procedure.
Obviously, X is a set containing all available path pairs
for r(s,d). If X=NULL, then reject the connection request.
Otherwise, only choose an optimal path pair (primary
path Pp and backup path Pb) with the minimal Cpb from
the set X to provision the service for current request
according to (9).
( , )
( , )
{ ( )}p
p b
pb T i E M ij ij
i j P
ij
i j P P
C P e P P z P
t
(9)
where tij is a binary variable. If one additional wavelength
is needed on link (i, j) to establish the connection, tij=1.
Otherwise, tij=0. β is a constant to adjust more attention
being paid on energy saving or on improvement of
resource usage. Different value of β means that (9) is
favorable to choose a pair of paths passing through links
whether with less additional resources or with less
additional energy consumption. A bigger value of β
means a higher degree of spare capacity sharing, and it is
favorable to improve the bandwidth utilization.
Fig. 2 shows the flow chart of the proposed DPA-SPP
algorithm. Its complexity is mainly determined by the
complexity of Yen’s K-shortest path algorithm and the
procedure of comparison operations and adjustments of
link cost, which is approximately O(K|N|3+K|L|). If K=1,
the complexity of Yen’s K-shortest algorithm is reduced
from O(K|N|3) to O(|N|2). Consequently, the complexity
of DPA-SPP is reduced to O(|N|2+|L|).
IV. SIMULATION RESULTS AND ANALYSIS
In this section, we will evaluate the performance of
DPA-SPP via extensive simulations by comparing it with
two other shared path protection algorithms, i.e., PU-SPP
and PA-SPP. PU-SPP mainly focuses on the sharing of
spare capacity, while PA-SPP stresses on the energy
saving with less concern on the network performance in
term of traditional metrics, such as blocking probability,
resource utilization, etc. In the rest of the section, we first
describe the test network topology and performance
metrics used in our simulations before presenting the
results.
Journal of Communications, Vol. 8, No.1, January 2013
60
Input the physical
topology information
Wait for a connection request
Connection setup request?
Modify the cost of links
according to Eqs. (2) and (3).
Compute K shortest paths by Yen's algorithm; All found
paths compose a set SK.
Sk =NULL?
Y
Figure 2. Flow chart of the proposed algorithm
l=1
Choose a path Pl from SK
l <K and at least one path is residual in SK?
l=l+1
X=NULL?
Put the path pair Pl
and Pc into set X.
Modify the cost of links
according to Eqs. (6), (7) and (8);
Compute the shortest path Pc by
Dijkstra's algorithm, which is
different from Pl , i.e., Pl≠Pc.
Found?
Y
Choose the optimal path pair with minimal cost Cpb from the set X to
create connection for current request.
Block the request
Y
Update the usage information of
links and nodes in the network.
Release the
connection and
update the usage
information of
links and nodes
in the network.
N
Y
N
N
N
Y
N
0
1
2
47
3
6
510
18
20158 11
9
13
12
16
14 19
22
21
17 23
800
1000
1000
250
800
1000
850
1000
1200
1150
1000
950
10001200
1900
1000
900850
950
1000
900
900
1300
1400
1000
600
1000
322
2600
650
1200
280
900
1000
900
850
600
800
1000
300
1300
700
1200
0
1
2
3
4
5
6
7
8
9
10
1000
410
340
390
200
750
270
310
370
400
600
900
730
320
600440
590850
810 720
710
350
710220
320
750
(a) (b)
Figure 3. Topology of test networks with fiber lengths (in km) marked on each link: (a) sample US network (USNET);
(b) Pan-European test network topology (COST 239).
A. Network Topology
The two test networks are the Pan-European test
network topology (COST 239) with 11 nodes and 26
bidirectional fiber links and a sample US network
topology (USNET) consisting of 24 nodes and 43
bidirectional fiber links [4] shown in Fig. 3, where nodes
are interconnected by bi-directional fiber links, and
physical distance of a fiber (in km) is marked on each
link. All nodes have wavelength conversion capabilities
and the number of wavelengths per fiber is assumed to be
12 and 15 for COST 239 and USNET, respectively. All
the traffic connection requests are bidirectional, which
are uniformly distributed among all node pairs, and there
is no knowledge about future requests. Each time there is
only one connection request. The arrival process of a
connection request is a Poisson process with arrival rate λ
and the connection holding time follows a negative
exponential distribution with mean 1 . If the algorithm
could not provision a reliable connection, the request is
rejected immediately without waiting queue. In the
simulations, let K=3, α=0.3, β=90 and =0.001, and the
total number of connection requests is generated up to 105.
The power consumption values of different devices are
assumed according to [8], [16], and [17], which are listed
in Table II.
Journal of Communications, Vol. 8, No.1, January 2013
61
TABLE I. POWER CONSUMPTION OF DIFFERENT DEVICES
Device Power consumption (W)
Electronic control system 150
Optical wavelength converters and 3D
MEMS switching per wavelength 1.757
Transponder 5.9
Pre-amplifier 10
Post-amplifier 20
In-line amplifier 15
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
30 40 50 60 70 80 90 100 110 120 130
Bac
kup
-pri
mar
y
ban
dw
idth
rati
o
Network Load (Erlang)
PU-SPP
PA-SPP
DPA-SPP
(a)
(b)
Figure 4. Backup-primary bandwidth ratio (BBR) versus network
offered load: (a) sample US network (USNET); (b) Pan-European test
network topology (COST 239).
B. Performance Metrics
The following performance metrics are used to
evaluate the three algorithms.
1) Backup-primary bandwidth ratio (BBR): BBR
represents the percentage of wavelengths reserved for
backup paths over the amount of wavelengths occupied
by primary paths, which can be written as follows.
( , )
( , )
iji j L
iji j L
r
BBRa
(10)
It is obvious that a smaller value of BBR means
smaller backup bandwidth reserved on all the backup
paths, and a higher degree of spare capacity sharing.
2) Blocking probability (BP): BP is defined as the ratio
between the number of blocked connection requests and
the number of all arriving connection requests during the
entire simulation period. In the case of dynamic traffic,
BP can approximately reflect the effectiveness of
bandwidth utilization. A smaller BP means higher
bandwidth utilization ratio, and vice versa.
3) Average power-saving ratio (APR): APR is defined
as the difference between the total power consumption of
a specific algorithm, i.e., DPA-SPP, PA-SPP and PU-SPP,
and the total power consumption of PU-SPP over the total
power consumption of PU-SPP, which can be written as
follows.
U S
U
E EAPR
E
(11)
where EU represents the total power consumption of PU-
SPP, and ES represents the total power consumption for a
specific algorithm, i.e., PA-SPP or DPA-SPP. A bigger
value of APR means higher energy saving, and vice versa.
Obviously, the value of APR for PU-SPP is always equal
to 0.
C. Simulation results
Fig. 4 shows the performance of PU-SPP, PA-SPP and
DPA-SPP in terms of backup-primary bandwidth ratio
(BBR) for different network load. We can observe that
PU-SPP performs best, followed by DPA-SPP and PA-
SPP in sequence. The reason for this is that PU-SPP
encourages picking a pair of link-disjoint paths with less
additional spare resources for all connection requests. On
the other side, PA-SPP and DPA-SPP pay more attention
to energy saving, especially PA-SPP only considering the
reduction of energy consumption while ignoring the
enhancement of resource utilization. Therefore, compared
with PU-SPP, more spare capacities are reserved in PA-
SPP and DPA-SPP, which leads to a bigger value of BPR.
With a joint consideration of energy saving and
improvement of resource usage, DPA-SPP has a higher
sharing of spare resources and resource utilization than
PA-SPP. Another observation from Fig. 4 is that the
values of BBR for the three algorithms are reduced with
an increase of network load. The reason for this is that
more network load means more connection requests
arriving at the network. Potentially, more connections are
established in the network. Therefore, spare capacities
can be shared among more connections, which can reduce
the value of BBR.
Fig. 5 compares the performance of the three
algorithms in terms of blocking probability (BP). In all
algorithms, BP is found to increase monotonically with
an increase of network load. PA-SPP yields the worst BP
and PU-SPP the best, with DPA-SPP falling in between.
The reason for this is that DPA-SPP and PU-SPP pay
attention to choose a pair of link-disjoint paths with less
additional resources for each connection request. It is
straightforward that both of them are favorable to reduce
the total amount of resources for provisioning a service
compared with PA-SPP that only considers power saving.
With more available resources, DPA_SPP and PU_SPP
can increase the chance to successfully establish
connections for connection requests arriving later. Being
different from PU-SPP that only emphasizes resource
utilization, DPA-SPP jointly considers improvement of
resource usage and energy saving. In order to put more
components into sleep state for the reduction of energy
consumption, it may choose a path with more hops but
with active components instead of a path with sleep or
idle components but with less hop as primary path, while
choose a path with sleep or idle components but with
Journal of Communications, Vol. 8, No.1, January 2013
62
more hops as backup path instead of a path with less hop
but with active components. This potentially increases the
total resources occupied by primary path and reserved for
backup path and leads to a bigger value of BP than PU-
SPP. Another observation from Fig. 5 is that under a light
network load (being less than 50 Erlang in USNET and
40 Erlang in COST 239), the BPs of PA-SPP and DPA-
SPP are very close to PU-SPP while DPA-SPP still
performs better than PA-SPP. However, with an increase
of network load, DPA-SPP is gradually far from PA-SPP
and close to PU-SPP. The reason for this is that many
free resources are available to create connections under
the case of light network load. Under this case, there is a
slight improvement for PU-SPP and DPA-SPP to have
more chance to establish connections successfully,
although both of them have a higher sharing of spare
capacity and resource utilization than PA-SPP. With an
increase of network load, resources gradually become the
major factor that determines whether a connection can be
created successfully. With the objective to enhance
resource utilization, PU-SPP and DPA-SPP can reduce
the resource occupation, and potentially can accept more
subsequent connection requests and reduce the blocking
probability. Being different from DPA-SPP, PA-SPP only
takes it into account to choose path with less energy
consumption without considering resource utilization.
Therefore, more resources are occupied for the
acceptance of the same number of connection requests;
and it has a worse performance of BP than DPA-SPP and
PU-SPP. This confirms that an algorithm only focusing
on power saving might lead to an unacceptable
performance degradation especially under a heavy
network load where the network is with a relatively
limited resource.
0
0.05
0.1
0.15
0.2
0.25
0.3
30 40 50 60 70 80 90 100 110 120 130
Blo
ckin
g p
rob
abili
ty
Network Load (Erlang)
PU-SPP
PA-SPP
DPA-SPP
(a)
(b)
Figure 5. Blocking probability (BP) versus network offered load: (a)
sample US network (USNET); (b) Pan-European test network topology
(COST 239).
0
0.05
0.1
0.15
0.2
0.25
30 40 50 60 70 80 90 100 110 120 130
Av
era
ge p
ow
er-
sav
ing
rati
o
Network Load (Erlang)
PU-SPP
PA-SPP
DPA-SPP
(a)
(b)
Figure 6. Average power-saving ratio (APR) versus network offered
load: (a) sample US network (USNET); (b) Pan-European test network
topology (COST 239).
Fig. 6 shows the performance of the three algorithms
in terms of average power-saving ratio (APR). Obviously,
considerable power savings (up to 20 percent for PA-SPP
and 16 percent for DPA-SPP in USNET, and up to 27
percent for PA-SPP and 22 percent for DPA-SPP in
COST 239, respectively) can be achievable. Another
observation from Fig. 6 is that with an increase of
network load, the value of APR for PA-SPP and DPA-
SPP first increase and then decrease. The reason for this
is that under a light network load, most of components
are in off or sleep mode. There is no significant
advantage for PA-SPP and DPA-SPP to reduce energy
consumption even if they deliberately pack primary paths
and backup paths on active components and sleep
components, respectively. With an increase of network
load, more connections are established in the network,
and more and more primary paths for new requests may
be packed on routes with active components in PA-SPP
and DPA-SPP, which is useful to reduce energy
consumption compared with PU-SPP. For PU-SPP with
the objective of resource utilization and without
distinguishing the component’s states, a primary path is
more possible to be carried by a route including idle or
sleep components, which are required to be activated with
more energy consumption. Therefore, the APRs of PA-
SPP and DPA-SPP gradually increase and reach the top
with an increase network load. However, under a case of
heavy network load, it is more possible that there are a
Journal of Communications, Vol. 8, No.1, January 2013
63
very few components being in off mode and a component
may be used by both primary path and backup path.
Under this case, it is difficult for PA-SPP and DPA-SPP
to pack primary path and backup path on different
components in different states. Thus, there is no
significant difference for PU-SPP, PA-SPP and DPA-SPP,
and the values of APR for PA-SPP and DPA-SPP are
gradually close to PU-SPP.
V. CONCLUSIONS
This paper proposes a dynamic power-aware shared
path protection algorithm (DPA-SPP) for WDM mesh
networks. DPA-SPP considers not only energy saving but
also resource utilization and spare capacity sharing. In
order to reduce power consumption, to improve sharing
of spare capacity, and to reduce blocking probability for
connection requests, DPA-SPP encourages to establish a
primary path for each connection request on active
resources, to pack backup paths on sleep resources, and to
leave more idle resources as far as possible while with a
consideration to prevent a link with very few number of
free wavelengths to become a bottleneck link, which is
beneficial to further improve the successful probability
for connection establishments. Under dynamic traffic
with different load, extensive simulations are performed
to compare the performance of our proposal with PU-SPP
and PA-SPP. Simulation results show that DPA-SPP can
combine the advantages of PU-SPP and PA-SPP to make
a tradeoff between resource utilization and energy saving.
It can achieve considerable performance gains leading to
reductions in spare capacities and average energy
consumption.
ACKNOWLEDGMENT
This work was supported in part by Fundamental
Research Funds for the Central Universities under grants
No. 2011QN022 and No. 2011JC026, NSFC under grant
No. 61171175, Liaoning BaiQianWan Talents Program
under grants No. 2009921063 and No. 2010921063, and
Open Research Fund of Key Lab of Optical Fiber Sensing
& Communications (UESTC), Ministry of Education.
REFERENCES
[1] Y. Zhang, P. Chowdhury, M. Tornatore, and B. Mukherjee,
“Energy efficiency in telecom optical networks,” IEEE Communi.
Surveys & Tutorials, vol. 12, no. 4, pp.441-457, 2010.
[2] A. Coiro, M. Listanti, A. Valenti, and F. Matera, “Reducing power
consumption in wavelength routed networks by selective switch
off of optical links,” IEEE J. of Selected Topics in Quantum
Electronics, vol. 17, no. 2, pp. 437-445, 2011.
[3] M. Xia, M. Tornatore, Y. Zhang, P. Chowdhury, C. U. Martel, and
B. Mukherjee, “Green provisioning for optical WDM networks,”
IEEE J. of Selected Topics in Quantum Electronics, vol. 17, no. 2,
pp. 437-445, 2011.
[4] A. Jirattigalachote, C. Cavdar, P. Monti, L. Wosinska, and A.
Tzanakaki, “Dynamic provisioning strategies for energy efficient
WDM networks with dedicated path protection,” Optical
Switching and Networking, vol. 8, no. 3, pp. 201–213, July 2011.
[5] Y. Wu, L. Chiaraviglio, M. Mellia, and F. Neri, “Power-aware
routing and wavelength assignment in optical networks,” in Proc.
of European Conference on Optical Communication, ECOC, 2009,
Vienna, Austria, September 2009, pp. 1-2.
[6] G. Shen and R. S. Tucker, “Energy-minimized design for IP over
WDM networks,” IEEE/OSA J. of Optical Communi. and
Network., vol. 1, no. 1, pp.176–186, June 2009.
[7] J. Baliga, R. Ayre, K. Hinton, W. V. Sorin, and R. S. Tucker,
“Energy consumption in optical IP networks,” IEEE/OSA Journal
of Lightwave Technology, vol. 27, no. 13, pp. 2391-2403, 2009.
[8] A. Coiro, M. Listanti, and A. Valenti, “Dynamic power-aware
routing and wavelength assignment for green WDM networks,” in
Proc of IEEE ICC, June 2011, pp. 1-6.
[9] A. Coiro, M. Listanti, A. Valenti, and F. Matera, “Power-aware
routing and wavelength assignment in multi-fiber optical
networks,” J. Opt. Commini. Netw., vol. 3, no. 11, pp. 816-829,
Nov. 2011.
[10] C. Cavdar, “Energy-efficient connection provisioning in WDM
optical networks,” in Proc. of IEEE/OSA Optical Fiber
Communication Conference and Exposition, OFC, 2011,
California, USA, March 2011, pp. 1-3.
[11] P. Wiatr, P. Monti, and L. Wosinska, “Green lightpath
provisioning in transparent WDM networks: Pros and cons,” in
Proc. of IEEE International Symposium on Advanced Networks
and Telecommunication Systems, ANTS, 2010, Mumbai, India,
December 2010, pp. 10-12.
[12] P. Wiatr, P. Monti, and L. Wosinska, “Power savings versus
network performance in dynamically provisioned WDM
networks,” IEEE Communi. Mag., vol. 50, no. 5, pp.48-55, May
2012.
[13] P. Ho and H. T. Mouftah, “Shared protection in mesh WDM
networks,” IEEE Communi. Mag., vol. 42, no. 1, pp. 70-76, Jan.
2004.
[14] R. He, B. Lin, L. Li, “Dynamic service-level-agreement aware
shared-path protection in WDM mesh networks,” J. of Network
and Computer Applications, vol. 30, no. 2, pp.429-444, April 2007.
[15] A. Muhammad, P. Monti, I. Cerutti, L. Wosinska, P. Castoldi, and
A. Tzanakaki, “Energy-efficient WDM network planning with
dedicated protection resources in sleep mode,” in Proc. of
GLOBECOM, Dec. 2010, pp.1-5.
[16] C. Cavdar, F. Buzluca, and L. Wosinska, “Energy-efficient design
of survivable WDM networks with shared backup,” in Proc. of
GLOBECOM, Dec. 2010, pp.1-5.
[17] S. S. Jalalinia, C. Cavdar, and L. Wosinska, “Survivable green
optical backbone networks with shared path protection,” in
Optical Fiber Communication Conference (OFC), Los Angeles,
California, March 2012, pp. OM2G.3.
[18] N. Bao, L. Li, H. Yu, Z. Zhang, and H. Luo, “Power-aware
provisioning strategy with shared path protection in optical WDM
networks,” Optical Fiber Techn., vol. 18, no. 2, pp.81-87, March
2012.
[19] G. Li, S. Wang, C. Kalmanek, and R. Doverspike, “Efficient
distributed restoration path selection for shared mesh restoration,”
IEEE/ACM Trans. Netw., vol. 11, no. 53, pp. 761-771, Oct. 200.
[20] J. Yen, “Finding the K shortest loopless paths in a network,”
Management Science, vol. 17, no. 11, pp. 712-716, July 1971. doi:
10.1287/mnsc.17.11.712.
Journal of Communications, Vol. 8, No.1, January 2013
64
Rongxi He received his B.S. and M.S. degrees
from Dalian Maritime University in 1992 and
1995, and his Ph.D. degree from University of
Electronic Science and Technology of China in
2002, all in communication and information
system.
From 2002 to 2004, he was a postdoctoral
fellow in Network and Communication Center
of Northeastern University. From 2006 to 2007,
he was a visiting scholar in Broadband
Communications Research Group of University of Waterloo, Canada.
He is currently a professor of Dalian Maritime University. His
research interests include optical networks, wireless networks and IP
quality of service.
He is senior member of China Computer Federation (CCF) and a
member of IEEE and ACM.
Bin Lin received his B.S., M.S. and Ph.D.
degrees from Dalian Maritime University in
1993, 1996 and 2006, respectively, all in
communication and information system.
He is currently an associate professor in
Dalian Maritime University. His research
interests include wireless communications,
signal processing and computer networks.
Journal of Communications, Vol. 8, No.1, January 2013
65